Deep learning, i.e, machine learning based on deep artificial neural networks, is one of the fastest growing fields of artificial intelligence research today, having outperformed competing methods in many areas of machine learning applications, e.g., image classification, face detection/recognition, natural language understanding and translation, speech recognition and synthesis, personal assistants (Siri, Google Now, Cortana), game-playing (e.g., Go, Poker), medical diagnosis, and self-driving vehicles. These deep artificial neural networks are able to capture complex nonlinear relationships using hierarchical internal representations which are learned automatically from the training data.
In the NCSA Gravity Group, we are applying deep learning with artificial neural networks, in combination with HPC numerical relativity simulations, in a variety of multimessenger astrophysics applications. Our current focus is on signal processing for gravitational wave detectors (LIGO, VIRGO, NANOGrav), analyzing data from telescopes (DES, LSST), and modeling waveforms from gravitational wave sources using algorithms that learn from numerical relativity simulations. This allows for real-time detection and parameter estimation of gravitational wave signals in LIGO, for denoising LIGO data contaminated with non-Gaussian noise, and for classification and unsupervised clustering of glitches (anomalies) in the LIGO detectors. We are now also developing fast automated transient search algorithms based on deep learning using raw image data from telescopes (e.g., DES and LSST) to rapidly classify electromagnetic counterparts to gravitational wave events.